Project Summary/Abstract Symptom-based diagnoses of mental illness are highly comorbid, biologically heterogeneous, and poorly predictive of treatment response. The National Institute of Mental Health has led efforts to redefine mental illness by its biological causes, establishing the Research Domain Criteria (RDoC) framework as a guide for investigating variation in basic brain systems. RDoC has been influential, named in hundreds of grants and publications, but it has yet to be systematically validated. It is unknown whether circuit-function links underlying the RDoC brain systems are reproducible across studies, and organizing principles remain largely untested. While the structure of RDoC as a modular hierarchy has evidence in resting state analyses, it has not been shown whether this applies to systems that support the diverse mental states affected in psychiatric disease. It is necessary to validate RDoC, and moreover, to establish fundamental principles of organization for systems defined jointly by human brain structure and function. The objective of this proposal is to apply large- scale computational neuroimaging meta-analyses to build a data-driven ontology that will not only serve as a benchmark in evaluating the validity of RDoC but also characterize the architecture of systems for human brain function. The long-term goal is to redefine mental illness by differences from healthy function within the brain systems of a data-driven ontology, facilitating rational targeting of neuromodulation treatments. The proposed meta-analyses will be the most comprehensive in the field with 18,155 MRI and PET studies already collected. The mental functions considered in these studies have been extracted from article texts using natural language processing, and brain circuits will be mapped from the brain coordinate data that were reported. The hypothesis is that brain systems are comprised of reproducible circuit-function links organized into a modular hierarchy, which for some systems will require updates to RDoC. This will be tested by comparing RDoC systems against those of a data-driven ontology. Aim 1: The reproducibility of circuit-function links will be evaluated by the performance of neural network classifiers predicting functions in article texts from circuits in brain scan data, and vice versa. Aim 2: The modularity of brain systems will be evaluated by a graph theoretic approach, and hierarchical structure will be assessed by representational similarity analysis. The impact of this project will be to validate the foremost psychiatry research framework and to characterize human brain systems through an innovative computational strategy. Together with targeted academic training in neurobiology, the fellowship is designed to offer preparation for a career as a physician-scientist leading advances in computational psychiatry. Training will be supported by an environment that combines world-class computing resources with esteemed and engaged mentors in psychiatry, neuroscience, and computer science.